What are the Durability and Susceptibility Metrics?

Jeremy Funk Posted On July 16, 2017

If you have followed my recent work, it is no surprise that I have been attempting to explain player injury profiles. I’ll be honest – injuries are mostly random events that are very difficult to predict in a statistical model with accuracy or reliability. While the recent remodel of sports injury predictor is a large step in the right direction, I place my faith in the building of metrics that seek to describe the player’s prior injury history.

My heart behind this project was simple: I want to wade into a topic that is ignored and misunderstood. While fantasy football has had several medical doctors provide insights on individual injuries, none have been able to apply combine both the medical and statistical perspective. Due to this void, I was recently commissioned to by sports injury predictor (SIP) in attempt to apply clinical relevance to a machine learning procedure. As an epidemiologist, one of the primary objectives was to use the duality of my professional training to build metrics that aids in the accuracy of SIP’s predictive models. This led me to the development of the following metrics: Durability Score and Susceptibility Score.

[am4show have=’g1;’ guest_error=’sub_message’ user_error=’sub_message’ ]

Durability Score

Durability score is an injury agnostic metric that identifies the player’s likelihood of suiting up on Sunday each fantasy season. In nature, this metric is both dynamic and descriptive. A dynamic metric was necessary as it allows for the score to change throughout a player’s career. A descriptive metric is a non-predictive value used to characterize qualitive traits. A great example of a descriptive metric is the Nike SPARQ Score, which seeks to outline the athleticism of a player based on combine results.

The heavily vetted database provided through my work in the revamping of sports injury predictor was my primary source of data. After applying the equation to each player (Durability Metric below), the distribution of the calculated scores was analyzed to establish five cut points to be used as durability groups. The name of this procedure is “Risk Stratification” and it is implemented as a simple way to visually compare scores with ease. Below, I have provided a few players from each of the durability risk groups to serve as an example.

Table 1

screen shot 2017 07 16 at 20.59.31

The players with the highest rates of playing with hold the largest durability score, while those who have missed a high rate of games due to injury will have a lower durability metric score. Generally, players within the top three groups (three through five) have displayed durability scores that should be targeted. On the other hand, the remaining two groups are filled with players who have shown an increased rate of missing games due to injury.

Durability groups one and two should cause concern for the viability in each season, but not panic. This assertion is particularly important for younger players, because the durability metric becomes increasingly accurate as players age. After the third year in the league, a player typically shows a large enough sample size to gather an idea of the player’s ability to see the field on Sundays.

While the durability metric is a large step towards characterizing player injury, it should not be viewed as a “silver bullet”. When used as a standalone, the durability metric is often misleading as it fails to provide the necessary context to differentiate between injury profiles. Specifically, the metric misses critical detail outlining high durability scores as an indicator of a player who is better at avoiding injury or better at playing despite an injury. The metric also showed similar questions for those with low durability, as the metric ignored those who are frequently injured and those effected by season-ending injuries. These shortcomings are what led me to develop the injury susceptibility metric.

Susceptibility Score

As stated above, the susceptibility metric was developed to help differentiate players who are constantly injured and those who have plagued by the misfortune few catastrophic injuries that ended their season early. To differentiate between these two classifications, it is important to understand two things: the proportion of unique injuries that explained a player’s missed games and the rate by which these injuries occurred.

Table 2

screen shot 2017 07 16 at 20.59.44

The multiplication of the susceptibility score by 100 was conducted to make interpretation of the scores understandable. The formal interpretation is this: At his current rate of injury, 6.25 injuries for Devonte Freeman would result in 100 games missed. After the calculation of each score, I applied the conducted the same risk stratification to the susceptibility metric as was done previously in the grouping of the durability scores. In the table above, it is clear the durability scores and susceptibility scores fail to mirror one another. As stated prior, the susceptibility score is intentionally built to provide additional context to the durability metric. Therefore, they must be used together in order to provide accurate insight.

Currently, the most noteworthy way to use these metrics is the identification of injury profiles with polar dichotomies in durability and susceptibility. In the example table above, Danny Woodhead returned a very low durability score (Durability Group=1) and a high susceptibility score (Susceptibility Group= 4). At first glance, it might be easy to identify Woodhead as a large injury risk each season and fail to see the whole picture. Further, Woodhead’s results indicate the proportion of his missed games were caused by few injuries. As stated previously, catastrophic injuries are often random events. Therefore similar injury profiles might work to identify players plagued by randomness. This holds especially true in younger players and will allow dynasty owners to identify the correct players to acquire at a discount.

On the other end of the spectrum, the susceptibility scores allow for the differentiation between those able to avoid injury from those who are likely to spend large portions of the season on the weekly injury report. Continuing to use the above example, Both Le’veon Bell and Devonta Freeman are found to have a low susceptibility of injury. A profile similar to Freeman’s (High Durability, Low Susceptibility) indicates a player who is largely able to avoid injury. These are the ideal players we seek to fill our rosters with. Specifically, In his four seasons in the league, Freeman has only missed one game on three unique injuries. This allows for his owners to trust that he is likely to be viable for the entire season. Though these scores are viable to change over time, Freeman has enough sample to expect him to avoid injury in both the short and long term.

For players like Le’veon Bell (Durability = 2, Susceptibility = 2), we begin to raise concerns over the short and long-term viability in a player. In this pairing archetype, we see a player who is particularly injury dense and unable to play through his ailments. While it is hard to advocate for the dismissal of such an incredible talent like Bell, it is important to understand that aging players rapidly decrease in durability while piling on a high number of injuries. This is particularly problematic in players on their first contract. Ignoring those players with small game samples, players displaying both low durability and susceptibility scores are players I would strongly avoid. In the coming weeks, I will be rolling out several articles identifying the injury profiles are encouraging, discouraging, and are confusing as hell.

[/am4show]